Software evolution from a time-series perspective

Abstract

Last decade of research on the software evolution field has brought successful empirical models for sofiware systems growth. Although the models are able to forecast the growth trend of a number of software systems, there still are some inconsistencies that must be addressed before the models can be incorporated into a final theory of software evolution. This study applies an innovative approach: the application of time series analysis techniques to historical data of software systems growth. Preliminary results have found the presence of anticorrelations in the data, phenomena that would validate some of the laws of software evolution.

Trend Analysis and Issue Prediction in Large-Scale Open Source Systems

“…One tool that is commonly used to characterize the evolution of software systems is time series [8,16,22,25]. In [25] the authors study the existence of long-range correlation in time series of software changes in order to investigate whether it represents the temporal signature of SelfOrganised Criticality.…”

“…They use the Rescaled Range Analysis technique formulated by Hurst in 1951 [14]. In [8] time series analysis techniques are applied to historical data of software systems growth to determine the process memory. The authors show the presence of long-term power-low anticorrelations in the data, which proves the existence of systems dynamics and thus validate some lows of software evolution.…”

On the prediction of the evolution of libre software projects

“…In 2002, Fuentetaja and Bagert [13] also explored the use of time series to obtain a model for the evolution of software projects. However they did not provide a model to predict the evolution but some tools which could be useful to obtain such a model.…”

Time series trends in software evolution

“…the still ongoing debate over whether some deterministic function captures the linear, sublinear, or super-linear characteristics behind software growth trends [6,7]. In addition to these growth curves, later research has approached software evolution time series from a different, stochastic, perspective [8][9][10]. In predictive modeling, the demarcation is easy: the best model wins.…”